Quantum Benchmark Zoo

← All benchmarks

QB GSEE Benchmark

Application-level

DARPA Quantum Benchmarking program's ground-state energy estimation benchmark: classical and quantum solvers are scored on solvability, accuracy, and runtime over a shared library of molecular Hamiltonians.

Measures
Solvability (probability of reaching chemical accuracy within runtime limits), accuracy, and runtime across a molecular Hamiltonian library, plus logical-qubit and T-gate estimates for quantum solvers
Introduced by
DARPA Quantum Benchmarking program (Bellonzi et al.: L3Harris, Zapata AI, University of Toronto, HRL & others) (2024)
Also known as
QB Ground State Energy Estimation Benchmark, QB-GSEE-Benchmark, QB-GSEE

The QB GSEE Benchmark scores classical and quantum solvers on the same concrete task: estimate molecular ground-state energies to chemical accuracy within runtime requirements. It was developed under DARPA’s Quantum Benchmarking (QB) program by a team spanning L3Harris, Zapata AI, the University of Toronto, HRL Laboratories, Boeing, and others (Bellonzi et al.), with the repository hosted by USC ISI, the program’s test-and-evaluation organization. “QB” is the program, not a vendor, and this benchmark is distinct from the same program’s BenchQ resource-estimation framework and from DARPA’s separate Quantum Benchmarking Initiative (QBI).

How it works

The benchmark ships a library of molecular electronic-structure Hamiltonians (228 instances with reference solutions, plus unsolved “guidestar” instances), following the instance-library pattern of HamLib and QOBLIB. The Hamiltonian FCIDUMP files sit on an L3Harris SFTP server (credentials required); metadata, schemas, scripts, and results live on GitHub. Solvers submit results and are scored on solvability, the probability of reaching chemical accuracy within the runtime requirement, along with accuracy and runtime. Quantum solvers additionally report estimated logical-qubit and T-gate utilization: today the quantum side is evaluated through logical resource estimates for double-factorized quantum phase estimation, not hardware runs. Planted-solution instances with known ground states (Wang et al.) keep hard instances checkable.

Strengths and limitations

Pitting quantum projections against strong classical baselines is the point, and the first results are sobering: the 2025 paper found fully optimized SHCI (classical) achieves near-universal solvability on the current instance set, so the benchmark does not yet expose a quantum-advantage frontier; the authors plan to add more strongly correlated instances. Quantum “results” remain projections rather than executions, and the credentialed SFTP hosting of instance files adds friction.

Notable results

The repository has been public since October 2024, with commits through June 2026, and the framework remains open to new solver submissions from classical and quantum developers. Note the dates: alpha releases (v0.1.0-alpha.1, December 2024; alpha.2, February 2025) preceded the August 2025 companion paper.

Key papers

Reference implementations

  • HamLib : More than 1.5 million pre-encoded qubit Hamiltonians (spin models, chemistry, and combinatorial optimization) supplying standardized problem instances for application-level quantum benchmarking.
  • QOBLIB : Open library of ten classically hard, practically motivated optimization problem classes (the Intractable Decathlon) with common reporting rules for quantum and classical solvers.